Dynamic hidden feature space detection of noisy image set by weight binarization

Historical documents are mostly in printed format. Considering space requirements and physical inspection, their preservation and restoration are costly. Scanners can turn these materials into an electronic mode, producing images polluted with noise. As a result, there is a higher storage demand and worse OCR precision. To overcome this, the most appropriate choice is noise reduction. The low-resolution grayscaled image and binarization process reduces the input data source. Furthermore, hidden feature space is extracted based on binary pixel quantization by the KF-CM method to obtain the feature space from binary images. The local-minimal points in binarized image segments define the 33 variables in the preprocessing stage. Followed by preprocessing, the scanned document images point KF-CM method is described as grouping input image pixels into noise, text, and background categories based on their characteristics. Therefore, noise reduction and binarization were both completed at the same time. The proposed approach has binarized a noisy image's bit planes by choosing local thresholds. This approach is evaluated with the document image datasets and compared with widely used binarization-based existing feature extraction methods, wherein the proposed work outperforms all other methods.